Seismic moment tensor event screening

SUMMARY Event screening is an explosion monitoring practice that aims to identify an event as an explosion (‘screened in’) or not (‘screened out’). Confidence in event screening can be increased if multiple independent approaches are used. We describe a new approach to event screening using the seismic moment tensor and its representation on the hypersphere, specifically the 5-sphere of 6-degree unit vectors representing the normalized symmetric moment tensor. The sample of moment tensors from an explosion data set is unimodal on the 5-sphere and can be parametrized by the Langevin distribution, which is sometimes referred to as the Normal distribution on the hypersphere. Screening is then accomplished by finding the angle from the explosion population mean to any newly measured moment tensor and testing if that angle is in the tail of the Langevin distribution (conservatively quantified as greater than 99.9 per cent of the cumulative density). We apply the screen to a sample of earthquakes from the Western USA and the September 2017 explosion and subsequent collapse at the Pungyye-Ri Test Site in North Korea. All the earthquakes and the collapse screen out, but the explosion does not.

State-Space Models on the Stiefel Manifold with a New Approach to Nonlinear Filtering

We develop novel multivariate state-space models wherein the latent states evolve on the Stiefel manifold and follow a conditional matrix Langevin distribution. The latent states correspond to time-varying reduced rank parameter matrices, like the loadings in dynamic factor models and the parameters of cointegrating relations in vector error-correction models. The corresponding nonlinear filtering algorithms are developed and evaluated by means of simulation experiments.